ETRI Journal, Volume 24, Number 2, April 2002 Seung Kwan Kim et al.

81
This paper describes our design of a hybrid amplifier
composed of a distributed Raman amplifier and erbium-
doped fiber amplifiers for C- and L-bands. We character-
ize the distributed Raman amplifier by numerical simula-
tion based on the experimentally measured Raman gain
coefficient of an ordinary single mode fiber transmission
line. In single channel amplification, the crosstalk caused
by double Rayleigh scattering was independent of signal
input power and simply given as a function of the Raman
gain. The double Rayleigh scattering induced power pen-
alty was less than 0.1 dB after 1000 km if the on-off Raman
gain was below 21 dB. For multiple channel amplification,
using commercially available pump laser diodes and fiber
components, we determined and optimized the conditions
of three-wavelength Raman pumping for an amplification
bandwidth of 32 nm for C-band and 34 nm for L-band. Af-
ter analyzing the conventional erbium-doped fiber ampli-
fier analysis in C-band, we estimated the performance of
the hybrid amplifier for long haul optical transmission.
Compared with erbium-doped fiber amplifiers, the optical
signal-to-noise ratio was calculated to be higher by more
than 3 dB in the optical link using the designed hybrid am-
plifier.

Manuscript received Sept. 21, 2001; revised Feb. 18, 2002.
Seung Kwan Kim (phone: +82 42 860 6143, e-mail: skim@etri.re.kr), Sun Hyok Chang (e-
mail: shchang@etri.re.kr), Jin Soo Han (e-mail: hanjinsoo@etri.re.kr), and Moo Jung Chu (e-
mail: mjc@etri.re.kr) are with Electronics and Telecommunications Research Institute, Daejeon,
Korea.
I. INTRODUCTION
As the demand for higher transmission capacity in wave-
length division multiplexing (WDM) systems increases, chan-
nel speed, channel number, and spectral efficiency need to be
upgraded. To this end, Raman amplifiers have become essential
in overcoming the limitations of the bandwidth, noise figure
(NF), and output power of conventional doped fiber amplifiers
[1]-[8]. Recently, great attention has been paid to the Raman
amplifiers used in conjunction with erbium-doped fiber amplifi-
ers (EDFAs) to form hybrid amplifiers, especially when the sys-
tem capacity needs to be upgraded by raising channel speed and
spectral efficiency without bandwidth expansion [3]-[8].
In the hybrid amplifiers, the NF can be effectively reduced
without changing the output power [1]. Both discrete Raman
[3], [4] and distributed Raman amplifiers (DRAs) [5]-[8] have
been used for hybrid amplifiers. For long haul transmission, a
DRA is more desirable since the transmission line itself can be
used as a gain medium simply by adding a Raman pump mod-
ule (RPM), and a previously installed EDFA can also be used
by making a slight modification according to the changed input
power and spectrum.
In designing hybrid DRA/EDFAs, it is important to optimize
the Raman gain in terms of the amplifier noise because the in-
crement of Raman gain increases double Rayleigh scattering
(DRS) noise as well as amplified spontaneous emission (ASE)
noise. In addition, the NF of the following EDFA also increases
because of the increased input power. Therefore, it is essential
to understand the behavior of hybrid DRA/EDFAs, which are
dependent on pump power, the corresponding Raman gain,
and the corresponding EDFA input.
This paper analyzes the characteristics of DRAs by numeri-
cal simulation based on the experimentally measured Raman
Design of Hybrid Optical Amplifiers for
High Capacity Optical Transmission
Seung Kwan Kim,

Sun Hyok Chang, Jin Soo Han, and Moo Jung Chu
82 Seung Kwan Kim et al. ETRI Journal, Volume 24, Number 2, April 2002
gain coefficient using an ordinary single mode fiber (SMF). We
evaluate the validity of the numerical simulation by comparing
it with the experimental measurement using a nonzero disper-
sion shifted fiber (NZ-DSF). In single channel amplification,
we examine the ASE and DRS noise behavior as the Raman
gain increases and then estimate the Raman gain limitation due
to the DRS noise induced penalty. In multiple channel amplifi-
cation, we determine the optimum pump wavelengths for a
bandwidth of 66 nm (32 nm for C-band and 34 nm for L-band)
and calculate the maximum Raman gain that is readily avail-
able with the current active and passive component technology.
We then determine by numerical simulation the optimum EDF
lengths for a C-band EDFA to produce the minimum noise fig-
ure. Finally, we estimate the performance of the resultant hy-
brid amplifier in C-band to show the applicability of the ampli-
fier to long haul optical transmission.
II. CONFIGURATION OF THE HYBRID
AMPLIFIER
A hybrid amplifier consists of a DRA and EDFAs as shown
in Fig. 1. A DRA is composed of a Raman pump module
(RPM) and a power clamping module (PCM) that keeps the
input power to the EDFA constant with a voltage controlled at-
tenuator (VCA) and a signal tapping coupler with a negative
feedback method. The DRA is pumped by several pump wave-

lengths in a backward direction to achieve a wide gain band-
width covering both C-band (15301562 nm) and L-band
(15711605 nm) with minimized pump induced signal fluctua-
tion [9]. Three or more pump wavelengths are multiplexed by a
pump WDM combiner (PWC) that is commercially available.
Each pump wavelength is provided by two pump laser diodes
(LDs) with the same wavelength multiplexed by a polarization
beam combiner (PBC). Each pump LD is wavelength stabi-
lized using an external fiber Bragg grating (FBG). The multi-
plexed pump beams are then launched into a transmission fiber
using a pump/signal WDM coupler (PSWC). The Raman am-
plified WDM signals are power regulated by the PCM and
passed to the C- and L-band EDFAs after C/L-band splitting.
Each EDFA has two amplification gain blocks (GBs) separated
by an inter-stage functional block for gain flattening and dis-
persion compensation. Each GB consists of EDF, a forward
(Fwd) and backward (Bwd) PSWC with optical isolators (OIs),
980 nm or 1480 nm pump LDs, and tap couplers for input and
output monitoring. The amplified WDM signals from the C-
and L-band EDFAs are then combined with a C/L-band com-
biner and conveyed to the next transmission fiber.
III. THEORETICAL MODEL OF THE DRA
The evolution of Raman pumps and WDM signals passing
through the transmission fiber is governed by (1) (3) [10], [11].


Fig. 1. Configuration of the hybrid optical amplifier. C, controller (proportional-integral contol); C-GB, C-band gain block; L-GB, L-band
gain block; C/L, C/L-band WDM coupler; DCM, dispersion compensating module; FC, fiber connector; GFF, gain flattening
filter; LD, pump laser diodes; PBC, polarization beam combiner; PWC, pump WDM combiner; PSWC, pump/signal WDM
coupler; PD, photo-diode; PCM, power clamping module; RPM, Raman pump module; Tap, tap coupler; VCA, voltage controlled
attenuator; VOA, variable optical attenuator.
C/L
LD(2)
PBC PBC PBC
PWC
VOA GFF
C-GB1
C-GB2
L-GB1 L-GB2
VCA
Tap
SMF(60120km)
DCM
PSWC
FC FC
FC input output
LD(1) LD(1) LD(2) LD(3) LD(3)
C
PCM
RPM C/L-EDFA
PD
C/L
ETRI Journal, Volume 24, Number 2, April 2002 Seung Kwan Kim et al. 83

+ =
i i i
N S P (1)
[ ]
[ ]

<

<
+

>
+

|
|
|
.
|

\
|

=
i j
j i
i j
i j
i h j
ij j
j i i R i
i j j
ij j
j i i R i
pol
i j j
ij
i j j R
pol i i
i
S
e
h
A
g
S P P
A
g
C
S P P
A
g
C S
dz
dS

1
1
1 2
) , (
) , (
) , (
kT
) (
m
m
m

(2)
[ ]
[ ]
[ ]

<

<
+
>

+

>
+

|
|
|
.
|

\
|

|
|
|
.
|

\
|

+ +

=
i h j
ij j
j i i R i
i j j
ij j
j i i R i
pol
h i j j
ij
i j j R
i j j
ij
i j j R
pol i i i i
i
N
e
h
A
g
N P P
A
g
C
e
h P P
A
g
N P P
A
g
C P N
dz
dN
i j
j i
i j
i j
i j
i j

1
1
1 2
) , (
) , (
1
1
1
) , (
) , (
kT
) (
kT
) (
m
m
m
m
(3)
To treat the Rayleigh back scattering as a noise source, we
separated the pure signal part, S
i
, and the noise part, N
i
, out of
the optical power of the i-th wavelength component. In (1)
(3), C
pol
is the polarization scrambling factor between the two
optical beams that is assumed to be 0.5 because each pump
beam has two orthogonal polarizations and the signal polariza-
tions are randomly allocated in front of the transmission fiber.
i
and
i
are the attenuation coefficient and Rayleigh back-
scattering coefficient of the i-th wavelength, respectively. A
ij
is
the effective interaction area defined in [12], which is equal to
the average value of the effective mode area of the two wave-
lengths if the Gaussian approximation of the fiber mode is used
as shown in (4) (6).
( )
(

|
.
|

\
|
= =
2
0 0
exp ) (

r
I r I r I
(4)
2
2 =
eff
A (5)
( ) ( )
( ) ( )
2
, , 2 2 j eff i eff
j i
A
j i
A
j
A
i
ij
A A
da r r
da r da r
A
+
= + =

(6)
The Raman gain coefficient (RGC), g
R
) v v (
j i
- ,
1
, is given as
a function of the frequency difference between the two wave-
lengths as well as a wavelength of light acting as a pump, be-
cause the peak Raman gain coefficient is scaled inversely pro-
portional to the pump wavelength [10]. The constant k and h
are Boltzmann's constant and Planck's constant, respectively.
The second term in the right hand side of (2) represents the
gain by optical signals with shorter wavelengths, while the
third and the fourth terms represent absorption by optical sig-
nals with longer wavelengths and the ASE generated by the
signal at the i-th wavelength, respectively. The second term in
the right hand side of (3) represents Rayleigh backscattering,
which is treated as a noise source, and the fourth term repre-
sents the addition of the ASE noise generated between the opti-
cal frequencies
i
v and v v
i
+ by optical signals with
shorter wavelengths. The other terms in (3) have the same
meaning as those in (2).
In order to theoretically analyze the DRA, the RGC needs to
be measured for a given optical fiber; to do this, we used four
single mode fiber (SMF) spools, each with a length of about 20
km. The fiber was a matched-cladding type SMF with an aver-
age mode field diameter of 9.37 m at 1310 nm and a cut-off
wavelength of 1270 nm. The RGC can be obtained by measur-
ing the small signal Raman gain (G) on a log scale with a given
pump power (P
p
), an effective interaction area (A
eff
) between
the pump and signal, an attenuation coefficient of the pump
) (
p
, and the fiber length (L) using (7) and (8) [10].
eff p pol
eff
s p p R s p p R
L P C
GA
g g
34 . 4
) ( ) ( ) , (
max ,
= =
(7)
,
) exp( 1
p
p
eff
L
L

= (8)
where g
R,max
) (
p

is the peak of the RGC with a pump wave-
length of ) (
p
and ) v v
s p
- ( is the normalized RGC by
g
R,max
. ) (
p

For a signal source we used a tunable laser diode with a
wavelength tuned by 1.0 nm from 1520.0 nm to 1620.0 nm
and an input power of 20 dBm. The pump wavelengths were
1445.7 nm and 1485.1 nm. With these wavelengths it was pos-
sible to measure the Raman gain in a range from 4.6 THz to
22.3 THz. To measure the Raman gain between 0 and 4.6 THz,
a new pump source was constructed at a wavelength of 1530.0
nm. This new pump source was a tunable laser diode amplified
by high power EDFA with a bandpass filter to reject unwanted
ASE noise [13]. Strictly speaking, A
eff
in (7) has a different
value for a different pump and signal wavelength combination.
84 Seung Kwan Kim et al. ETRI Journal, Volume 24, Number 2, April 2002
However, it differs by only 0.28 dB for signal wavelengths
from 1520.0 nm to 1620.0 nm, which is less than the level flat-
ness of the optical spectrum analyzer (OSA) of 0.2 dB. There-
fore, the same A
eff
estimated at the Raman peak wavelength
was used for each pump wavelength. For 1445.7 nm and
1485.1 nm, the effective interaction areas were 84.8 m
2
and
89.1 m
2
, respectively. For the 1530.0 nm pump, the Raman
gain peak was located outside of the measurement range, and
the measurement data was used only to connect the gap of the
line shape function ) v v
s p
- ( , between 0 and 4.6 THz. To
minimize errors in the measurement, the Raman gain was
measured for 20, 40, 60, and 80 km, and the RGC was aver-
aged over the fiber length. For the 1445.7 nm and 1485.1 nm
pump, the gain peak wavelengths were 1545.0 nm and 1590.0
nm resulting in a Raman frequency shift of 13.3 THz. The av-
erage peak values of the measured RGC divided by A
eff
were
0.8810
-3
m
-1
W
-1
for the 1445.7 nm pump and 0.8010
-3

m
-1
W
-1
for the 1485.1 nm pump. Since the RGC is scaled in-
versely proportional to the pump wavelength, the scalar prod-
uct of the RGC and the pump wavelength is an important pa-
rameter for simulation using arbitrary multiple pump wave-
lengths. From the two average peak values, the estimated aver-
age value of g
R,max
p p
) (
for the two pump wavelengths was
1.0710
-19
m
2
/W with less than 1.2 % error. This was much
less than the overall measurement error caused by the level
flatness of the OSA and the polarization dependent loss (PDL)
of the fiber components. Figure 2 shows the measured Raman
gain coefficient from 0 to 22.3 THz normalized by its maxi-
mum value.
For the computer simulation, the measured attenuation coef-
ficients of the pump and signal wavelengths were used and the
Rayleigh backscattering coefficient was taken from Ref. [14],


Fig. 2. Measured Raman gain coefficient normalized by the
maximum value.
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
0 2 4 6 8 10 12 14 16 18 20 22 24
Frequency shift (THz)
N
o
r
m
a
l
i
z
e
d

R
a
m
a
n

g
a
i
n

c
o
e
f
f
i
c
i
e
n
t


assuming that it was not significantly different. In solving (1)
(3), the Runge-Kutta method with a 4th order was applied with
a relaxation method to find correct solutions for all signals. The
length was divided by 50 and it was proved that the error was
less than 0.01 dB compared with a length division of 100. It
took less than 3 min. to calculate 3 pumps and 160 channels us-
ing a Pentium III 933 MHz microprocessor.
IV. EVALUATION OF THE NUMERICAL
SIMULATION
Before moving into the characterization and design of the
DRA, we examined the validity of the numerical simulation
using nonzero dispersion shifted fiber (NZ-DSF), where the
RGC divided by A
eff
was 1.510
-3
and 1.310
-3
m
-1
W
-1
for the
1445.7 and 1485.1 nm pumps, respectively. The average cut-
off wavelength and mode field diameter of the NZ-DSF was
1069 nm and 8.25 m, respectively. The calculated effective
interaction areas were 49.2 and 52.7 m
2
for pump-signal
combinations of 1445.7 nm to 1545.0 nm and 1485.1 nm to
1590.0 nm, respectively. The estimated average value of
g
R,max
p

was the same as that of the SMF, that is, 1.110
-19

m
2
/W with less than 2% error for the two pump-signal combi-
nations. As with the SMF, the normalized RGC profile for the
NZ-DSF was also obtained with the two pump wavelengths
and the 1530.0 nm pump source.
When the single channel input of 20 dBm tuned from 1520
to 1620 nm was launched into NZ-DSF with a length of 80 km,
the on-off Raman gain and the equivalent NF were measured
and are shown as symbols in Fig. 3, where the pump power at
1485.1 nm was varied from 80 mW to 110 mW while it was
fixed at 135 mW at 1445.7 nm. The numerical simulation re-
sults shown as solid lines agreed well with the experimental
data. The maximum difference between the experiment and
numerical simulation was less than 0.3 dB for the gain and 0.5
dB for the NF; these are comparable to the error bounds that
are due to PDL, the wavelength uniformity of the PSWC, and
the level flatness of the OSA.
For the measurement of multiple channel amplification, 24
continuous-wave channels in C-band and 22 in L-band with
two intensity modulated link control channels were launched
into the 80 km NZ-DSF as shown in Fig. 4. The total input
powers were 18.7 dBm for C-band and 19.3 dBm for L-band.
The off-level in Fig. 5 corresponded to the estimated output
power when the SRS effect was ignored; it was calculated us-
ing the measured fiber attenuation coefficient. Without pump-
ing, the channels became tilted because of the SRS effect.
When the pump power at 1485.1 nm was varied from 20 mW
to 100 mW with a fixed pump power of 135 mW at 1445.7 nm,
ETRI Journal, Volume 24, Number 2, April 2002 Seung Kwan Kim et al. 85

Fig. 3. Small signal on-off Raman gain and NF spectrum of DRA using NZDSF 80 km for various Raman pump powers. The numbers in
the legend box correspond to pump power at 1485.1 nm, while the pump power at 1445.7 nm was fixed at 135 mW. n.sim.,
numerical simulation.
-3
-2
-1
0
1
2
3
4
5
6
7
8
9
1520 1530 1540 1550 1560 1570 1580 1590 1600 1610
Wavelength (nm)
O
n
-
o
f
f

g
a
i
n

(
d
B
)

-2.5
-2.0
-1.5
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
E
q
u
i
v
a
l
e
n
t

N
F

(
d
B
)

80mW
90mW
100mW
110mW
n.sim.
n.sim.
n.sim.
n.sim.
80mW
90mW
100mW
110mW
n.sim.
n.sim.
n.sim.
n.sim.


Fig. 4. Input spectrum for multiple channel amplification measurement of DRA using NZDSF.
-60
-50
-40
-30
-20
-10
0
10
20
1525 1535 1545 1555 1565 1575 1585 1595 1605
Wavelength (nm)
O
p
t
i
c
a
l

p
o
w
e
r

(
d
B
m
)

Link control channel
24 channels 22 channels
18.7 dBm 19.3 dBm
Res: 0.1 nm

the output power profiles changed as shown in Fig. 5. Again, the
experimental data agreed reasonably well with the numerical
simulation. However, the measured NF data could not be com-
pared with the numerical simulation because they exhibited
enormous errors owing to the high input ASE level.

V. SINGLE CHANNEL AMPLIFICATION
For the single channel amplification, 1450 nm and 1550
nmwere chosen as the pump and signal wavelengths, respec-
tively. The bandwidth was set at 0.1 nm. The fiber length was
86 Seung Kwan Kim et al. ETRI Journal, Volume 24, Number 2, April 2002

Fig. 5. Output spectrum of DRA using NZDSF 80 km for various pump powers. The numbers in the legend box correspond to pump power
at 1485.1 nm, while the pump power at 1445.7 nm was fixed at 165 mW. exp, experiment; n.sim., numerical simulation.
-25
-20
-15
-10
-5
0
1525 1535 1545 1555 1565 1575 1585 1595 1605
Wavelength (nm)
O
u
t
p
u
t

p
o
w
e
r

(
d
B
m
)

exp(w/o pump)
exp(20mW)
exp(60mW)
exp(100mW)
n.sim.(w/o pump)
n.sim.(20mW)
n.sim.(60mW)
n.sim.(100mW)
off level


Fig. 6. On-off Raman gain and equivalent NF versus pump power for various input power levels without DRS. The numbers in the legend
box correspond to input power levels in dBm.
0
5
10
15
20
25
30
35
40
45
50
0 200 400 600 800 1000 1200 1400
Pump power (mW)
O
n
-
o
f
f

g
a
i
n

(
d
B
)

Pump: 1450 nm
Signal: 1550 nm
Without DRS noise
-6
-5
-4
-3
-2
-1
0
1
E
q
u
i
v
a
l
e
n
t

N
F

(
d
B
)

G(-20)
G(-15)
G(-10)
G(-5)
G(0)
G(5)
NF(-20)
NF(-15)
NF(-10)
NF(-5)
NF(0)
NF(5)

assumed to be 80 km. Because the increasing rate of gain was
less than that of the ASE at the beginning of the pump power,
the OSNR decreased as the pump power increased (Fig. 7) until
the on-off Raman gain became about 9 dB when the pump
power was about 300 mW (Fig. 6). After that, the OSNR in-
creased monotonically. If DRS was ignored, the noise power

generated from the ASE was almost independent of the input
signal power until the pump power reached 800 mW at which
the on-off Raman gain was about 24 dB. Hence, the OSNR
(P
sig
/P
ASE
) increased by an equal amount of signal power incre-
ment (Fig. 7). Since the attenuation of the SMF was 17.1 dB,
including both input and output connectors, it could be made
ETRI Journal, Volume 24, Number 2, April 2002 Seung Kwan Kim et al. 87

0
10
20
30
40
50
60
70
0 200 400 600 800 1000 1200 1400
Pump power (mW)
P
o
w
e
r

r
a
t
i
o

(
d
B
)

Pin= 5 dBm
-20 dBm
0 dBm
-15 dBm
-10 dBm
-5 dBm
BW: 0.1 nm
Pump: 1450 nm
Signal: 1550 nm
Psig/(PASE+PDRS)
Psig/PASE(W/O DRS)
Psig/PASE(W/ DRS)
Fig. 7. OSNR versus pump power with and without DRS noise for
various input power levels. Thick solid lines, narrow solid
lines, and dotted lines indicate P
sig
/(P
ASE
+P
DRS
), P
sig
/P
ASE

with DRS, and P
sig
/P
ASE
without DRS, respectively.

lossless when the pump power was higher than 600 mW (Fig. 6).
The equivalent noise figure at a pump power of 600 mW was
about 2 dB.
When the DRS of both the signal and the ASE was taken
into account, the noise characteristics changed. In this case, the
optical noise power included both the ASE power (P
ASE
) and

the DRS power of the signal (P
DRS
), where the ASE power
consisted of the ASE power generated by the Raman process
and the DRS of the ASE. In Fig. 7, the OSNR, which is the
power ratio of the output signal to optical noise
(P
sig
/(P
ASE
+P
DRS
)), decreased again as the pump power in-
creased and there were local maxima for input powers lower
than 0 dBm. For higher input power, the OSNR degraded
monotonically. P
DRS
was proportional to the signal input power
at low pump powers and the amount was not comparable to
P
ASE
unless the input power was larger than 0 dBm. However,
P
DRS
grew faster than the signal gain as the pump power in-
creased and it eventually exceeded P
ASE
. Hence, the OSNR
limitation appeared at a much lower pump power range than
when only the DRS of the ASE was considered (Fig. 7), which
was the case examined by Hansen et al [15]. When the
crosstalk from the DRS of the signal was calculated as a func-
tion of on-off Raman gain for various input powers, the
crosstalk was uniquely determined by gain, regardless of the
input power (Fig. 8). Since the power penalty due to DRS of a
signal is given by (9) [16],
,
2
1
1 log 10
2
10
|
|
.
|

\
|
= NQ
P
P
sig
DRS
I
(9)
where N and Q are the number of spans and system Q, respec-
tively, the crosstalk that gives a power penalty of 0.1 dB after13
span (> 1000 km) at a bit error ratio of 10
-12
(Q=7.03) was cal-
culated to be 41.5 dB corresponding to an on-off gain of


Fig. 8. DRS of signal induced crosstalk versus on-off Raman gain.
0
10
20
30
40
50
60
0 5 10 15 20 25 30 35 40 45 50
On-off gain (dB)
C
r
o
s
s

t
a
l
k

(
d
B
)

5 dBm
0 dBm
-5 dBm
-10 dBm
-15 dBm
Pump: 1450 nm
Signal: 1550 nm
88 Seung Kwan Kim et al. ETRI Journal, Volume 24, Number 2, April 2002
about 21 dB. Therefore, the DRS noise can be ignored when
the Raman gain is below 21 dB, which corresponds to a pump
power of about 700 mW from Fig. 6.
VI. MULTIPLE CHANNEL AMPLIFICATION
For multiple channel amplification, our aim was to accom-
modate 160 channels in both C-band (1530.01562.0 nm) and
L-band (1571.01605.0 nm) with channel spacing of 50 GHz,
where 80 channels were equally allocated to each band. Since
the wavelength shift of the peak Raman gain from the pump
wavelength is about 100 nm in the 1500 nm region, 1445 nm
and 1485 nm can be the simplest combination of the Raman
pump wavelengths. In a DRA, however, the pump with the
shorter wavelength transfers its energy not only to WDM sig-
nal channels but also to the pump with the longer wavelength
as it passes through a long length of fiber. Consequently, a two
or three times larger pump power is normally required for the
shorter wavelength pump than the longer wavelength pump in
order to uniformly amplify both the C- and L-bands. Therefore,
it is necessary to allocate two pump wavelengths for C-band
amplification when pump LDs with finite pump power are to
be used. When a commercial PWC capable of multiplexing 4
pump wavelengths with equal spacing is considered, roughly
five possibilities can be listed in selecting two pump wave-
lengths for C-band and one for L-band, where the shortest
pump wavelength is kept less than or equal to 1430 nm in order
to minimize OSNR degradation due to the gain tilt arising from
the SRS effect between signal channels. Cases A, B, and C
have wavelength spacing of 25 nm while cases D and E have
20 nm (Table 1). In order to determine the optimum pump
wavelength combination, the pump power of the shortest
wavelength was fixed at a maximum available value. To our
knowledge, currently the maximum power of the fiber cou-
pled output from a single wavelength stabilized LD with a fiber
Bragg grating (FBG) is 230 mW. The maximum insertion
losses of the PBC, PWC, and PSWC (Fig. 1) are 0.4, 0.8, and

Table 1. Five different Raman pump wavelength combinations.
Ch 1 (nm) Ch 2 (nm) Ch 3 (nm) Ch 4 (nm)
A 1420 1445 (1470) 1495
B 1425 1450 (1475) 1500
C 1430 1455 (1480) 1505
D 1425 1445 (1465) 1485
E 1430 1450 (1470) 1490


0.4 dB, respectively, so that the overall insertion loss before
launching into the SMF should be 2 dB including the splice
losses. Therefore, the output power of the two polarization
multiplexed LDs becomes 290 mW at the SMF input. In the
numerical simulation, input signal power per channel was as-
sumed to be 1.0 dBm, so that the total 160 channel input would
be 23 dBm.
At first, the pump powers of the first and second wavelength
were maximized. Then, the pump power of the third wave-
length was increased until the L-band output power became
equal to the C-band. If the average gain per band was below 10
dB, they were considered to be optimum pump power values.
If it exceeded 10 dB, the pump power of the second wave-
length was lowered until it got close to 10 dB and the output
power of each band was balanced using the third wavelength.
The resultant optimum pump powers and the corresponding
gains and output powers are summarized in Table 2. In cases A,
B, and C, the average gains did not reach 10 dB and the output
power excursions were larger than those in cases D and E (Fig.
9, 10). In cases D and E, 10 dB gains were achievable with the
pump powers at a second wavelength of above 250 mW and
240 mW for D and E, respectively. Case D had lower output
power excursion than case E, but case E had the advantage
when the extension of the gain bandwidth to the longer wave-
length side was considered in L-band. Between these two fac

Table 2. Optimum pump powers for the 2nd and 3rd pump wavelengths for each pump wavelength combination at maximum pump power of
the 1st pump wavelength, and the corresponding gain, output power, and output power excursion.
Ppump,2nd (mW) Ppump,3rd (mW) Gon-off,avg (C) (dB) Gon-off,avg (L) (dB) Pout (C) (dBm) Pout (L) (dBm) Pout (C) (dB) Pout (L) (dB)
A 290 108 9.39 9.57 12.2 12.2 2.63 2.45
B 290 122 9.72 9.93 12.6 12.6 2.25 2.77
C 290 122 9.72 9.87 12.5 12.5 2.53 2.43
D 250 160 10.17 10.33 13.0 13.0 1.71 1.90
E 240 160 10.18 10.33 13.0 13.0 2.28 2.00

ETRI Journal, Volume 24, Number 2, April 2002 Seung Kwan Kim et al. 89

-10
-9
-8
-7
-6
-5
-4
-3
1520 1530 1540 1550 1560 1570 1580 1590 1600 1610
Wavelength (nm)
O
u
t
p
u
t

p
o
w
e
r

(
d
B
m
)

A
B
C
Fig. 9. Output power spectrum of 160 channels for various pump
wavelength combinations. See Table 1 for meaning of A,
B, and C.


Fig. 10. Output power spectrum of 160 channels for various pump
wavelength combinations. See Table 1 for meaning of D
and E. The numbers in parentheses correspond to pump
powers for successive wavelengths in mW.
Wavelength (nm)
O
u
t
p
u
t

p
o
w
e
r

(
d
B
m
)

-9
-8
-7
-6
-5
-4
-3
-2
1520 1530 1540 1550 1560 1570 1580 1590 1600 1610
D
E (290-240-160)
E (290-290-177)

tors, the output power excursion or gain flatness of the Raman
amplifier itself is of less importance in the hybrid amplifier de-
sign because the quantity is small compared to that of the
EDFA, and it can be easily compensated for by a gain flatten-
ing filter (GFF) inserted between the GB1 and GB2 of the
EDFA. Consequently, case E was chosen for the optimum
pump wavelength combination for the DRA.
For case E, we examined the DRA characteristics as we in-
creased the pump power of 1450 nm and maximized that of
1430 nm while we correspondingly adjusted that of 1490 nm
so that both C- and L-band output power would be equal. As
the pump power of 1450 nm increased, the adjusted pump
power of 1490 nm increased linearly and so did the output
power in the log scale (Fig. 11). The output power excursion in
the log scale was also linearly proportional to the 1450 nm
pump power (Fig. 12). As we expected from the previous single

Fig. 11. Required pump power of 1490 nm pump versus pump
power of 1450 nm when the pump power of 1430 nm was
fixed at maximum. The second Y-axis correspond to total
output power in dBm and average on-off Raman gain for
C- and L-band in dB scale.
1490 nm pump Total output C-gain L-gain
90
100
110
120
130
140
150
160
170
180
190
190 200 210 220 230 240 250 260 270 280 290 300
1450 nm pump power (mW)
1
4
9
0

n
m

p
u
m
p

p
o
w
e
r

(
m
W
)

8
9
10
11
12
13
14
15
16
17
18
O
p
t
i
c
a
l

p
o
w
e
r

(
d
B
m
)
,

G
o
n
-
o
ff
,
a
v
g

(
d
B
)



Fig. 12. Output power excursion versus pump power of 1450 nm.
1430 nm and 1490 nm pump conditions were the same as
in Fig. 10.
1.8
1.9
2.0
2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9
3.0
230 240 250 260 270 280 290 300
1450 nm pump power (mW)
O
u
t
p
u
t

p
o
w
e
r

e
x
c
u
r
s
i
o
n

(
d
B
)

C-band
L-band

channel analysis, the DRS in this case did not cause any effect,
because the gain was less than 12 dB at best. The OSNR in-
crement due to DRS was no greater than 0.04 dB with a 0.1 nm
filter bandwidth. The NF ranged from 2.2 to +2.2 dB depend-
ing on the wavelengths as plotted in Fig. 13. Owing to the 1490
nm pump and SRS effect, the NF profile has a negative slope
in the wavelength domain [17]. Figure 14 shows the power
evolution of the 3 backward pumps and 8 sampled channels for
each band, where C-band channels experienced higher attenua-
tion than L-band channels before amplification. As mentioned
earlier in this section, the energy transfer from the shorter
wavelength pumps to the longer wavelength pump
90 Seung Kwan Kim et al. ETRI Journal, Volume 24, Number 2, April 2002
Fig. 13. On-off Raman gain and NF spectrum for two pump power
combinations.
290-240-160mW 290-290-177mW
290-240-160mW 290-290-177mW
0
2
4
6
8
10
12
14
1520 1530 1540 1550 1560 1570 1580 1590 1600 1610
Wavelength (nm)
O
n
-
o
f
f

g
a
i
n

(
d
B
)

-3
-2
-1
0
1
2
3
4
5
E
q
u
i
v
a
l
e
n
t

N
F

(
d
B
)



Fig. 14. Evolution of signal and pump power. For signal evolution,
8 channels were sampled for each band with 5 nm spacing
including the first and the last channels.
-14
-12
-10
-8
-6
-4
-2
0
2
0 10 20 30 40 50 60 70 80
Distance (km)
S
i
g
n
a
l

p
o
w
e
r

(
d
B
m
)

-10
-5
0
5
10
15
20
25
30
P
u
m
p

p
o
w
e
r

(
d
B
m
)

C-band
L-band
1430 nm
1490 nm
1450 nm

occurred because of the SRS effect (Fig. 14), which explains
why only 23 percent of the total launched pump power was re-
quired at the pump wavelength of 1490 nm in order to balance
the output power of C- and L-bands.
VII. ERBIUM-DOPED FIBER AMPLIFIER
The calculated output from the DRA was then sent to the
GB1 of the following EDFAs. In order to characterize the per-
formance of the hybrid amplifier theoretically, the C-band
EDFA was simulated exemplarily. Table 3 shows the parame-
ters of the EDF used for simulation. In the EDFA simulation,
Table 3. Simulation parameters for EDFA.
Parameter Value Tolerance
Cut-off wavelength 895 nm 5 %
Loss at 1558.5 nm 1.95 dB/m 5 %
Core radius 1.47 m 5 %
Refractive index difference (n) 0.0184 5 %
Numerical aperture 0.233 5 %
Peak absorption (near 1.53 m) 5.53 dB/m 5 %
Peak absorption (near 980 nm) 4.05 dB/m 5 %
MFD (Petermann 2) 1.55 m 5.75 m 5 %
Background loss 1.2 m 4.08 dB/km 5 %
Estimated background loss 1.55 m 1.58 dB/km 5 %
Fiber outside diameter 125 m 2 m
Estimated erbium concentration 8.110
24
m
-3





Fig. 15. Measured spectrum of absorption and gain per unit length
of the used erbium-doped fiber.
6
1450 1470 1490 1510 1530 1550 1570 1590 1610 1630 1650
Wavelength (nm)
900 920 940 960 980 1000 1020 1040 1060 1080 1100
980 nm absorption
absorption
gain
-1
0
1
2
3
4
5
G
a
i
n

a
n
d

a
b
s
o
r
p
t
i
o
n

(
d
B
/
m
)


we used the equation shown in Ref. [18], [19] with measured
absorption and emission data (Fig. 15) supplied by the manu-
facturer. The modal intensity profile was assumed to be Gaus-
sian. Since the input power of GB1 was higher than 5 dBm, in-
cluding the insertion loss of the PCM and C/L-band splitter,
both forward and backward pumping with full available power
at 980 nm in GB1 was assumed, that is 180 mW. To maximize
the output power, the EDF of GB2 was assumed to be pumped
by the forward and backward 1480 nm pump with an output
power of 200 mW. The PSWC for combining the pump and
signal is assumed to be a WDM/Tap Coupler/Isolator hybrid
type for forward pumping and a WDM/Isolator hybrid type for
backward pumping. The insertion loss of the PSWC was set to
1.4, 0.8, 0.8, and 0.8 dB for 980 nm forward, 980 nm backward,
ETRI Journal, Volume 24, Number 2, April 2002 Seung Kwan Kim et al. 91
1480 nm forward, and 1480 nm backward, respectively, with
signal ports, and 0.6 dB for all pumps with pump ports, in ac-
cordance with the data sheet of commercial products. The
splice loss between the 980 nm WDM coupler and the EDF
was about 0.2 dB for both 980 nm and 1550 nm. For the 1480
nm WDM coupler, it was about 0.8 dB for both the pump and
signal wavelengths. Therefore, the overall insertion loss for the
signal band was assumed to be 2.0, 1.9, 2.6, 2.5 dB for GB1
input, GB1 output, GB2 input, and GB2 output, respectively,
including the input and output fiber connectors, additional tap
couplers for backward pumping, and some margin for estima-
tion errors. The overall insertion loss for the pump band was
assumed to be 0.8, 0.8, 1.4, and 1.4 dB, respectively.
Between GB1 and GB2, the inter-stage containing the VOA,
GFF, and DCM was considered with a maximum insertion loss
of 16.5 dB assumed. A simulation program calculated the filter
profile of the inter-stage to obtain flattened output for all chan-
nels. After the total output power and the gain profile of GB2
were calculated with a flat input profile, the required input pro-
file that produced the flat output profile with the same total
output power negligibly changed the gain profile. The gain dif-
ference was less than 0.01 dB. Therefore, the desired input pro-
file was obtained from the flat output profile producing the
same output power divided by the gain profile calculated using
the flat input profile. The filter profile of the inter-stage could
then be calculated by comparing the output profile of GB1 and
the input profile of GB2. Once the inter-stage filter profile was
obtained, the gain, NF, and the output power of GB2 were cal-
culated again using the changed input profile to confirm the

correct result.
When the average gain and NF of GB1 over 80 channels
was calculated as a function of the EDF length, the maximum
output power of 20.66, 20.74, and 20.86 dBm could be ob-
tained at a length of 18, 1618, 1416 m, for the input power
of 0, 5, and 10 dBm, respectively. After these optimum lengths
were reached, output power decreased because the unpumped
region began to appear. There was a tendency for the optimum
length to be shorter for higher input power. The higher input
power caused faster decay of the pump, so the unpumped re-
gion began to appear at a shorter EDF length. The NF was an
increasing function of the EDF length. If the EDF length was
increased with a fixed pump power, the average population in-
version decreased and the spontaneous emission factor in-
creased to give a higher NF. Figure 16 shows the average gain
and NF as a function of input power when the length of the
EDF was 10, 12, and 14 m, respectively. The minimum NF of
5.1 dB was located at an input power of between 6 and 7
dBm for an EDF length of 10m. It corresponded to an internal
NF, of the EDF itself, of 3.1 dB, which is close to the limit of a
discrete amplifier. The input power for the local minimum NF
moved to a higher value, about 4 dBm for 16 m, as the EDF
length increased. After a 5 dBm input, the NF increased steeply.
It is possible to obtain an output power of more than 20 dBm as
long as the input power is greater than 0 dBm with an EDF
length longer than 10 m.
Unlike GB1, the output power of GB2 had no local maxi-
mum until the EDF length reached 22 m. This was because the
1480 nm pump had a slower absorption rate than the 980 nm


Fig. 16. Average gain and NF versus input power for various EDF lengths in GB1.
16 m
10 m
12 m
14 m
16 m
10 m
12 m
14 m
0
5
10
15
20
25
30
35
40
45
-40 -35 -30 -25 -20 -15 -10 -5 0 5 10 15
Total input power (dBm)
A
v
e
r
a
g
e

g
a
i
n

(
d
B
)

5.0
5.2
5.4
5.6
5.8
6.0
6.2
6.4
6.6
6.8
A
v
e
r
a
g
e

N
F

(
d
B
)

92 Seung Kwan Kim et al. ETRI Journal, Volume 24, Number 2, April 2002

Fig. 17. Average gain and NF versus input power for various EDF lengths in GB2.
0
5
10
15
20
25
30
35
40
45
-40 -35 -30 -25 -20 -15 -10 -5 0 5 10 15
Total input power
A
v
e
r
a
g
e

g
a
i
n

(
d
B
)

6.8
7.0
7.2
7.4
7.6
7.8
8.0
8.2
8.4
A
v
e
r
a
g
e

N
F

(
d
B
)

18 m
20 m
22 m
18 m
20 m
22 m

pump. The EDF length necessary to achieve more than 21
dBm output power was 20 m for 0 dBm input and 18 m for 5
dBm input. Since the maximum output power from GB1 was
about 20 dBm or more, the input power to GB2 was at least
about 3.5 dBm if the inter-stage loss of 16.5 dB was considered.
Therefore, the length of EDF for GB2 should be longer than 18
m. Figure 17 shows the average gain and NF of GB2 as a func-
tion of input power. The average NF is basically larger than that
of GB1 because the average inversion is low.
The total NF profiles of the EDFA (GB1+GB2) was then
calculated for several EDF length combinations including the
about 1 dB insertion loss of the C/L-band WDM splitter and
combiner when the fiber transmission loss was assumed to be
22 dB so that the input power might be 2 dBm. The increment
of the EDF length from 10 to 14m in GB1 uniformly reduced
the NF, but the effect was negligible after 14 m. However, the
EDF length change in GB2 produced rotation in the NF profile
with the pivot point at about 1546 nm (Fig. 18), and it rotated
counterclockwise as the length increased. The flattest NF pro-
file can be achieved with a length of 17 m. However, the out-
put power did not reach 20 dBm for 17 m, so that the opti-
mum length design was determined to be 14 m for GB1 and
18 m for GB2.
VIII. PERFORMANCE OF THE HYBRID
AMPLIFIER
Assuming that the input power per channel, gain, and NF of
the optical boosting amplifier (OBA) were 20 dBm, 21 dB,
and 6 dB, respectively, we estimated the OSNR trace of the cas-
caded amplifier without Raman pumping as shown in Fig. 19.


Fig. 18. NF spectrum of EDFA for various EDF length in GB2
when the EDF length of GB1 was 14 m and the Raman
pump was off.
6.0
6.5
7.0
7.5
8.0
8.5
9.0
1525 1530 1535 1540 1545 1550 1555 1560 1565
Wavelength (nm)
N
o
i
s
e

f
i
g
u
r
e

(
d
B
)

14 m, 16 m
14 m, 17 m
14 m, 18 m
14 m, 20 m
14 m, 22 m
Input: -2 dBm
GB1: 980 nm Fwd (180 mW), 980 nm Bwd (180 mW)
GB2: 1480 nm Fwd (200 mW), 1480 nm Bwd (200 mW)
Inter-stage loss: 16.5 dB

The wavelength dependence of fiber loss and the SRS effect
were not taken into account in the calculation. Because the NF
profile had the minimum and maximum values depending on
the wavelength in Fig. 18, the OSNR trace also had the mini-
mum and maximum values although the gain flatness was
less than 0.003 dB. After 10 spans, the OSNR became 21.2
22.2 dB.
For the Raman pumping, the input power of the EDFA was
set at 5 dBm using the PCM function and considering the desi-
gn margin and the additional minimum insertion loss of the
PSWC and PCM, which was assumed to be about 2 dB. Then,
the fiber transmission loss could be effectively reduced to 15
ETRI Journal, Volume 24, Number 2, April 2002 Seung Kwan Kim et al. 93
Fig. 19. OSNR traces for optical link with span loss of 22 dB. The
upper four traces correspond to minimum and maximum
OSNR depending on wavelength for hybrid amplifier,
while the lower traces for EDFA.
G=10.7dB
G=10.7dB
G=7.0 dB
G=7.0 dB
G=0 dB
G=0 dB
16
18
20
22
24
26
28
30
32
34
1 2 3 4 5 6 7 8 9 10 11 12 13 14
Number of amplifier
O
S
N
R

(
d
B
)

Hybrid amplifier
EDFA
OBA
G=21 dB
NF=6 dB
Pin = -20 dBm/channel

dB. Since the input profile to the EDFA was not flat due to the
Raman gain profile, the gain of the EDFA had inverted profiles
of Raman gain to produce a flat output power profile (Fig. 20).
With this high input power, the NF of the EDFA ranged from
about 8 dB to about 13 dB depending on the EDF length. The
effect of a change in EDF length was similar to the case of 2
dBm input.
In order to find the optimum length of EDF in the hybrid
amplifier design, the equivalent amplifier concept was used.
Using the same measured attenuation coefficient, the fiber
length was increased so that the average fiber attenuation over
the wavelength range would be equal to 22 dB. The output
from the fiber, which was 2 dBm, was conceptually thought
to be the input of the hybrid amplifier. The hybrid amplifier
was treated as a discrete amplifier with an on-off gain and
equivalent NF. Both the wavelength dependent fiber attenua-
tion and SRS effect were included in the on-off Raman gain to
compare the result with the case without the Raman amplifier.
Then the NF of the hybrid amplifier was between 4.1 and 5.3
dB (Fig. 21). Since the NF of the Raman amplifier had a nega-
tive slope (Fig. 13), the total NF became flat with an EDF
length of 20 m for GB2 because the slope was well compen-
sated for by the positive slope of the NF of the EDFA shown in
Fig. 20. The optimum EDF length for GB1 seemed to be 14 m
but the effect was less than 0.2 dB compared with 10 m in
terms of NF. The optimum EDF length for GB2 was deter-
mined to be 20 m from Fig. 21. Figure 22 shows the calculated

Fig. 20. Gain and NF spectrum of EDFA for various EDF length
in GB2 when the EDF length of GB1 was 12 m and the
Raman pump was on.
12 m, 18 m
12 m, 20 m
12 m, 22 m
12 m, 18 m
12 m, 20 m
12 m, 22 m
10
11
12
13
14
15
16
17
18
1525 1530 1535 1540 1545 1550 1555 1560 1565
Wavelength (nm)
G
a
i
n

(
d
B
)

8
9
10
11
12
13
N
F

(
d
B
)



Fig. 21. Total equivalent NF of hybrid amplifiers for various EDF
length in GB2 with EDF length in GB1 of 14 m.
14 m, 18 m 14 m, 20 m 14 m, 22 m
3.0
3.5
4.0
4.5
5.0
5.5
6.0
1525 1530 1535 1540 1545 1550 1555 1560 1565
Wavelength (nm)
E
q
u
i
v
a
l
e
n
t

N
F

(
d
B
)

14 m, 18 m 14 m, 20 m
14 m, 22 m

filter profile of the inter-stage, in which the background loss
for DCM and VOA was subtracted. It seems that the case of
10 m for GB1 and 20 m for GB2 is more desirable than the
other two in fabricating the gain flattening filter.
In order to evaluate the validity of the simulation result, the
GFF profile estimated by simulation was compared with the
experimentally measured GFF profile (Fig. 23). The GFF pro-
file was for the EDFA only, where the EDF length was 18 m
for GB1 and 20 m for GB2. The main difference came from
the fact that the simulated GFF profile included the wave-
length dependent loss of the DCM whereas the measured GFF
profile excluded the DCM. When the DCM loss profile was
94 Seung Kwan Kim et al. ETRI Journal, Volume 24, Number 2, April 2002
Fig. 22. Estimated inter-stage filter profiles to produce flat output
of EDFA.
10 m, 20 m 12 m, 20 m 14 m, 20 m
-9
-8
-7
-6
-5
-4
-3
-2
-1
0
1530 1535 1540 1545 1550 1555 1560 1565
Wavelength (nm)
A
t
t
e
n
u
a
t
i
o
n

(
d
B
)

12 m, 20 m
14 m, 20 m
10 m, 20 m

Fig. 23. Comparison of GFF profiles between simulation and
experiment for EDFA with EDF length for GB1 and
GB2 of 18 m and 20 m, respectively.
-12
-10
-8
-6
-4
-2
0
1530 1535 1540 1545 1550 1555 1560 1565
Wavelength (nm)
A
t
t
e
n
u
a
t
i
o
n

(
d
B
)

-11.4
-9.4
-7.4
-5.4
-3.4
-1.4
0.6
A
t
t
e
n
u
a
t
i
o
n

(
d
B
)

Experiment
DCM
Simulation

separately measured and drawn together (Fig. 23), the discrep-
ancy was compensated for.
The OSNR trace of the cascaded hybrid amplifier was esti-
mated as shown in Fig. 19. In comparison with the EDFA
without a DRA, the average OSNR was improved by 2.5 dB
after 10 spans when the Raman gain was 7.0 dB, including the
insertion loss of the PSWC and PCM. This was improved by
3.2 dB over the EDFA without a DRA when the Raman gain
was raised to 10.7 dB by increasing the 1450 nm pump fully
and re-adjusting the 1490 nm pump correspondingly, at which
time the PCM was removed and the GFF profile was reshaped
while the EDF length remained unchanged. The insertion loss
of the PSWC was assumed to be 1.0 dB. The equivalent NF in
this case was reduced to 3.44.5 dB. Figure 24 shows how the

Fig. 24. Average equivalent NF versus average on-off Raman gain
and the corresponding input power to EDFA.
3.0
4.0
5.0
6.0
7.0
8.0
-1 0 1 2 3 4 5 6 7 8 9 10 11 12 13
Average on-off Raman gain (dB)
A
v
e
r
a
g
e

e
q
u
i
v
a
l
e
n
t

N
F

(
d
B
)

-4
-2
0
2
4
6
8
10
E
D
F
A

i
n
p
u
t

(
d
B
m
)

EDFA only
EDFA only

average equivalent NF changed as the Raman gain increased. It
is clear that the increasing Raman gain was advantageous in
reducing the overall NF, although the NF of the EDFA in-
creased as the input power grew (Fig. 16). This fact can be ex-
plained by (10), which is derived from the NF equation of cas-
caded optical amplifiers [18],
,
|
.
|

\
|

+ =
DRA
DRA
EDFA
EDFA
DRA
EDFA
DRA
G
G
A NF
NF
A G
A NF
NF NF
(10)
where A is the attenuation between the DRA and EDFA, which
was 1.0 dB. Since NF
DRA
is negative and (NF
EDFA
A) is posi-
tive, (10) is negative if the parenthesis is negative, which ap-
plies when the increasing rate of the NF of the EDFA is smaller
than the DRA gain increasing rate. In Fig. 25, the average NF
of the EDFA increased by about 0.6 dB for a 1 dB increment of
the average DRA on-off gain.
VIV. CONCLUSIONS
Our study examined the characteristics of the DRA and
EDFA in order to achieve the optimum design for hybrid opti-
cal amplifiers. Our results showed that the DRS noise induced
power penalty became negligible after 13 spans of 80 km SMF
as long as the on-off Raman gain of the DRA was less than 21
dB. We determined that the three optimum pump wavelengths
for an amplification bandwidth of 32 nm for C-band and 34 nm
for L-band were 1430, 1450, and 1490 nm. Using commer-
cially available fiber optic components with highest pump LDs,
the on-off gain of the DRA comprising those three pump wave-
lengths was more than 11 dB. The total equivalent NF of the
hybrid amplifier decreased as the gain of the DRA increased as
long as the rate of increase of the NF for the EDFA was smaller
ETRI Journal, Volume 24, Number 2, April 2002 Seung Kwan Kim et al. 95

Fig. 25. Maximum, minimum, and average NF for DRA and EDFA,
respectively, as a function of on-off Raman gain.
max(DRA)
min(DRA)
avg(DRA)
max(EDFA)
min(EDFA)
avg(EDFA)
-2.0
0.0
2.0
4.0
6.0
8.0
10.0
12.0
14.0
16.0
18.0
5 6 7 8 9 10 11 12 13
DRA on-off gain (dB)
N
F

(
d
B
)

EDFA (GB1+GB2)
DRA

than the rate of increase of the gain for the DRA. For the fiber
optic transmission link with a 22 dB span loss, our hybrid am-
plifier improved the OSNR after 10 spans by 3.2 dB over the
EDFA without a DRA.
REFERENCES
[1] T.N. Nielson, Raman Amplifiers in WDM Systems, 12th LEOS
Annual Meeting, vol. 2, 1999, pp. 471-472.
[2] T. Tsuzaki, M. Kakui, M. Hirano, M. Onishi, Y. Nakai, and M. Ni-
shimura, Broadband Discrete Fiber Raman Amplifier with High
Differential Gain Operating over 1.65 m-Band, Optical Fiber
Comm. Conf. (Anaheim, CA, 2001), MA3.
[3] S. Kawai, H. Masuda, K. Suzuki, and K. Aida, Wide-Bandwidth
and Long-Distance WDM Transmission Using Highly Gain-
Flattened Hybrid Amplifier, IEEE Photon. Technol. Lett., vol. 11,
no. 7, 1999, pp. 886-888.
[4] W.Y. Oh, S.S. Lee, H. Lee, and W. Seo, 16-Channel C-Band Hy-
brid Fiber Amplifier Comprising an EDFA and a Single Diode
Laser Pumped Dispersion Compensating Raman Amplifier,
European Conf. On Comm. (Munich, Germany, 2000), 4.4.5.
[5] H. Suzuki, J. Kani, H. Masuda, N. Takachio, K. Iwatsuki, Y. Tada,
and M. Sumida, 1-Tb/s (10010Gb/s) Super-Dense WDM
Transmission with 25-GHz Channel Spacing in the Zero-
Dispersion Region Employing Distributed Raman Amplification
Technology, IEEE Photon. Technol. Lett., vol. 12, no. 7, 2000, pp.
903-905.
[6] Y. Zhu, W.S. Lee, C. Scahill, C. Fludger, D. Watley, M. Jones, J.
Homan, B. Shaw, and A. Hadjifotiou, 1.28Tbit/s (3240Gbit/s)
Transmission over 1000 km NDSF Employing Distributed Ra-
man Amplification and Active Gain Flattening, Electron. Lett.,
vol. 37, no. 1, 2001, pp. 43-45.
[7] B. Zhu, P.B. Hansen, L. Leng, S. Stulz, T.N. Nielsen, C. Doerr, A.J.
Stentz, D.S. Vengsarkar, Z.J. Chen, D.W. Peckham, and L.
Gruner-Nielsen, 800 Gbit/s (8010.66Gbit/s) Transmission over
3200 km of Non-Zero Dispersion Shifted Fibre with 100 km Am-
plified Spans Employing Distributed Raman Amplification,
Electron. Lett., vol. 36, no. 22, 2000, pp. 1860-1861.
[8] T.N. Nielsen, A.J. Stentz, K. Rottwitt, D.S. Vengsarkar, Z.J. Chen,
P.B. Hansen, J.H. Park, K.S. Feder, S. Cabot, S. Stulz, D.W.
Peckham, L. Hsu, C.K. Kan, A.F. Judy, S.Y. Park, L.E. Nelson,
and L. Gruner-Nielson, 3.28-Tb/s Transmission over 3100 km
Nonzero-Dispersion Fiber Using Dual C- and L-Band Distributed
Raman Amplification, IEEE Photon. Technol. Lett., vol. 12, no. 8,
2000, pp. 1079-1081.
[9] J.S. Wey, D.L. Butler, M.F. Leeuwen, L.G. Joneckis, and J. Gold-
har, Crosstalk Bandwidth in Backward Pumped Fiber Raman
Amplifiers, IEEE Photon. Technol. Lett., vol. 11, no. 11, 1999, pp.
1417-1419.
[10] G.P. Agrawal, Nonlinear Fiber Optics, 3rd ed., Academic Press,
San Diego, 2001, pp. 298-303.
[11] B. Min, W.J. Lee, and N. Park, Efficient Formulation of Raman
Amplifier Propagation Equations with Average Power Analysis,
IEEE Photon. Technol. Lett., vol. 12, no. 11, 2000, pp. 1486- 1488.
[12] R.H. Stolen, Nonlinear Properties of Optical Fibers, in Optical
Fiber Telecommunications, S.E. Miller and A.G. Chynoweth, ed.
Academic Press, San Diego, 1979, p. 130.
[13] B.H. Choi, M.J. Chu, H.H. Park, and J.H. Lee, Performances of
Erbium-Doped Fiber Amplifier Using 1530 nm-Band Pump for
Long Wavelength Multichannel Amplification, ETRI J., vol. 23,
no. 1, 2001, pp. 1-8.
[14] H. Kidorf, K. Rottwitt, M. Nissov, M. Ma, and E. Rabarijaona,
Pump Interactions in a 100-nm Bandwidth Raman Amplifier,
IEEE Photon. Technol. Lett., vol. 11, no. 5, 1999, pp. 530-532.
[15] P.B. Hansen, L. Eskildsen, A.J. Stentz, T.A. Strasser, J. Judkins, J.J.
DeMarco, R. Pedrazzani, and D.J. DiGiovanni, Rayleigh Scatter-
ing Limitations in Distributed Raman Pre-Amplifiers, IEEE Pho-
ton. Technol. Lett., vol. 10, no. 1, 1998, pp. 159-161.
[16] M. Nissov, K. Rottwitt, H.D. Kidorf, and M.X. Ma, Rayleigh
Crosstalk in Long Cascades of Distributed Unsaturated Raman
Amplifiers, Electron. Lett., vol. 35, no. 12, 1999, pp. 997-998.
[17] C.R.S. Fludger, V. Handerek, and R.J. Mears, Fundamental
Noise Limits in Broadband Raman Amplifiers, Optical Fiber
Comm. Conf. (Anaheim, CA, 2001), MA5.
[18] E. Desurvire, Erbium-Doped Fiber Amplifiers: Principles and
Applications, John Wiley & Sons, New York, 1994, Ch. 1, Ch. 2.
[19] T.G. Hodgkinson, Improved Average Power Analysis Technique
for Erbium-Doped Fiber Amplifiers, IEEE Photon. Technol. Lett.,
vol. 4, no. 11, 1992, pp. 1273-1275.






96 Seung Kwan Kim et al. ETRI Journal, Volume 24, Number 2, April 2002
Seung Kwan Kim received his BS degree in
physics from Seoul National University, Seoul,
Korea, in 1991. He received MS and PhD de-
grees in physics from KAIST, Daejeon, Korea,
in 1993 and 1997, respectively. From 1997 to
1998, he visited the Optoelectronics Division at
the University of Strathclyde in Glasgow, Scot-
land, where he worked on tunable fiber lasers.
In 1999, he joined ETRI, Daejeon, Korea, where he has been involved
in research and development on multiwavelength fiber lasers, high
power EDFAs, and fiber Raman amplifiers for high capacity WDM
transmission system. Dr. Kim is a member of OSK and OSA.

Sun Hyok Chang received his BS degree in
physics from Korea Advanced Institute of Sci-
ence and Technology (KAIST) in 1994. He re-
ceived MS and PhD degrees in physics from
KAIST in 1996 and 2000, respectively. He
joined Electronics and Telecommunications Re-
search Institute in 2000 and has been engaged in
research on WDM fiber optic telecommunica-
tion systems. He is currently interested in fiber Raman amplifiers and
nonlinear effects in the transmission fibers. Dr. Chang is a member of
the Optical Society of Korea (OSK).

Jin Soo Han received his BE degree in elec-
tronic engineering from Yonsei University,
Seoul, Korea, in 1998. He received his ME de-
gree from Korea Advanced Institute of Science
and Technology, Daejeon, Korea, in 2000, and
he joined Electronics and Telecomm- unications
Research Institute, Korea. His resear- ch inter-
ests are optical communication system, WDM
transmission technology, and optical network.

Moo Jung Chu received his BS degree in
physics from Yonsei University, Seoul, Korea,
in 1980. He received MS and PhD degrees from
Korea Advanced Institute of Science and Tech-
nology, Daejeon, Korea, in 1982 and 1985, re-
spectively. In 1985, he joined Electronics and
Telecommunications Research Institute (ETRI),
Korea. He is a Principal Member of Optical
Communications Department, ETRI. He has been engaged in research
of optical devices, optical amplifiers, WDM optical transmission tech-
nology. Dr. Chu is a member of KITE, KICS, OSK, and Korea Physi-
cal Society.